On Wed 05-06-19 21:25:01, Bharath Vedartham wrote: > IMO the reason why a lot of failures must not have occured in the past > might be because the programs which use it use stuff like cpu_to_node or > have checks for nid. > If one day we do get a program which passes an invalid node id without > VM_BUG_ON enabled, it might get weird. It will blow up on a NULL NODE_DATA and it will be quite obvious what that was so I wouldn't lose any sleep over that. I do not think we have any directly user controlable way to provide a completely ad-hoc numa node for an allocation. -- Michal Hocko SUSE Labs
